Skeleton Sequences Research Articles

An Efficient Feature Fusion of Graph Convolutional Networks and Its Application for Real-Time Traffic Control Gestures Recognition

Recently, skeleton-based gesture and action recognition have emerged thanks to the progress in human pose estimation. Gesture representation using skeletal data is robust since skeletal data are invariant to the individual’s appearance. Among different approaches proposed for skeleton-based action/gesture recognition, Graph Convolutional Network (GCN) and its variations have obtained great attention thanks to its ability to capture the graph essence of the skeletal data. In this paper, we aim to design an efficient scheme using relative joints of skeleton sequences adapted in a GCN framework. Both spatial features (i.e., joint positions) and temporal ones (i.e., the velocity of joints) are combined to form the input of Attention-enhanced Adaptive GCN (AAGCN). The proposed framework deals with limitations of the original AAGCN when it works on challenging datasets with incomplete and noisy skeletal data. Extensive experiments are carried out on three datasets CMDFALL, MICA-Action3D, NTU-RGBD. Experimental results show that the proposed method achieves superior performance compared with existing methods. Moreover, to illustrate the application of the proposed method in real-time traffic control gesture recognition for autonomous vehicles, we have evaluated the proposed method on the TCG dataset. The obtained results show that the proposed method offers real-time computation capability and good recognition results. These results suggest a promising solution to deploy a real-time and robust recognition technique for gesture-based traffic control in autonomous vehicles.

Pose recognition in sports scenes based on deep learning skeleton sequence model

Human skeleton extraction is a basic problem in the field of computer vision. With the rapid progress of science and technology, it has become a hot issue in the field of target detection such as pedestrian recognition, behavior monitoring, and pedestrian gesture recognition. In recent years, due to the development of deep neural networks, modeling of human joints in acquired images has made progress in skeleton extraction. However, most models have low modeling accuracy, poor real-time performance, and poor model availability. problem. Aiming at the above-mentioned human target detection problem, this paper uses the deep learning skeleton sequence model gesture recognition method in sports scenes to study, aiming to provide a gesture recognition method with strong noise resistance, good real-time performance and accurate model. This article uses motion video frame images to train the VGG16 network. Using the network to extract skeleton information can strengthen the posture feature expression, and use HOG for feature extraction, and use the Adam algorithm to optimize the network to extract more posture features, thereby improving the posture of the network Recognition accuracy. Then adjust the hyperparameters and network structure of the basic network according to the training results, and obtain the key poses in the sports scene through the final classifier.

A Novel Graph Representation for Skeleton-based Action Recognition

Graph convolutional networks (GCNs) have been proven to be effective for processing structured data, so that it can effectively capture the features of related nodes and improve the performance of model. More attention is paid to employing GCN in Skeleton-Based action recognition. But there are some challenges with the existing methods based on GCNs. First, the consistency of temporal and spatial features is ignored due to extracting features node by node and frame by frame. We design a generic representation of skeleton sequences for action recognition and propose a novel model called Temporal Graph Networks (TGN), which can obtain spatiotemporal features simultaneously. Secondly, the adjacency matrix of graph describing the relation of joints are mostly depended on the physical connection between joints. We propose a multi-scale graph strategy to appropriately describe the relations between joints in skeleton graph, which adopts a full-scale graph, part-scale graph and core-scale graph to capture the local features of each joint and the contour features of important joints. Extensive experiments are conducted on two large datasets including NTU RGB+D and Kinetics Skeleton. And the experiments results show that TGN with our graph strategy outperforms other state-of-the-art methods.

Adversarial Attack on Skeleton-Based Human Action Recognition

Deep learning models achieve impressive performance for skeleton-based human action recognition. Graph convolutional networks (GCNs) are particularly suitable for this task due to the graph-structured nature of skeleton data. However, the robustness of these models to adversarial attacks remains largely unexplored due to their complex spatiotemporal nature that must represent sparse and discrete skeleton joints. This work presents the first adversarial attack on skeleton-based action recognition with GCNs. The proposed targeted attack, termed constrained iterative attack for skeleton actions (CIASA), perturbs joint locations in an action sequence such that the resulting adversarial sequence preserves the temporal coherence, spatial integrity, and the anthropomorphic plausibility of the skeletons. CIASA achieves this feat by satisfying multiple physical constraints and employing spatial skeleton realignments for the perturbed skeletons along with regularization of the adversarial skeletons with generative networks. We also explore the possibility of semantically imperceptible localized attacks with CIASA and succeed in fooling the state-of-the-art skeleton action recognition models with high confidence. CIASA perturbations show high transferability in black-box settings. We also show that the perturbed skeleton sequences are able to induce adversarial behavior in the RGB videos created with computer graphics. A comprehensive evaluation with NTU and Kinetics data sets ascertains the effectiveness of CIASA for graph-based skeleton action recognition and reveals the imminent threat to the spatiotemporal deep learning tasks in general.

MM-Fit

Fitness tracking devices have risen in popularity in recent years, but limitations in terms of their accuracy and failure to track many common exercises presents a need for improved fitness tracking solutions. This work proposes a multimodal deep learning approach to leverage multiple data sources for robust and accurate activity segmentation, exercise recognition and repetition counting. For this, we introduce the MM-Fit dataset; a substantial collection of inertial sensor data from smartphones, smartwatches and earbuds worn by participants while performing full-body workouts, and time-synchronised multi-viewpoint RGB-D video, with 2D and 3D pose estimates. We establish a strong baseline for activity segmentation and exercise recognition on the MM-Fit dataset, and demonstrate the effectiveness of our CNN-based architecture at extracting modality-specific spatial temporal features from inertial sensor and skeleton sequence data. We compare the performance of unimodal and multimodal models for activity recognition across a number of sensing devices and modalities. Furthermore, we demonstrate the effectiveness of multimodal deep learning at learning cross-modal representations for activity recognition, which achieves 96% accuracy across all sensing modalities on unseen subjects in the MM-Fit dataset; 94% using data from the smartwatch only; 85% from the smartphone only; and 82% on data from the earbud device. We strengthen single-device performance by using the zeroing-out training strategy, which phases out the other sensing modalities. Finally, we implement and evaluate a strong repetition counting baseline on our MM-Fit dataset. Collectively, these tasks contribute to recognising, segmenting and timing exercise and non-exercise activities for automatic exercise logging.

A Novel Graph Representation for Skeleton-based Action Recognition

Graph convolutional networks (GCNs) have been proven to be effective for processing structured data, so that it can effectively capture the features of related nodes and improve the performance of model. More attention is paid to employing GCN in Skeleton-Based action recognition. But there are some challenges with the existing methods based on GCNs. First, the consistency of temporal and spatial features is ignored due to extracting features node by node and frame by frame. We design a generic representation of skeleton sequences for action recognition and propose a novel model called Temporal Graph Networks (TGN), which can obtain spatiotemporal features simultaneously. Secondly, the adjacency matrix of graph describing the relation of joints are mostly depended on the physical connection between joints. We propose a multi-scale graph strategy to appropriately describe the relations between joints in skeleton graph, which adopts a full-scale graph, part-scale graph and core-scale graph to capture the local features of each joint and the contour features of important joints. Extensive experiments are conducted on two large datasets including NTU RGB+D and Kinetics Skeleton. And the experiments results show that TGN with our graph strategy outperforms other state-of-the-art methods.

Sparse Adaptive Graph Convolutional Network for Leg Agility Assessment in Parkinson's Disease.

Motor disorder is a typical symptom of Parkinson's disease (PD). Neurologists assess the severity of PD motor symptoms using the clinical rating scale, i.e., MDS-UPDRS. However, this assessment method is time-consuming and easily affected by the perception difference of assessors. In the recent outbreak of coronavirus disease 2019, telemedicine for PD has become extremely urgent for clinical practice. To solve these problems, we developed an automated and objective assessment method of the leg agility task in the MDS-UPDRS using videos and a graph neural network. In this study, a sparse adaptive graph convolutional network (SA-GCN) was proposed to achieve fine-grained quantitative assessment of skeleton sequences extracted from videos. Specifically, the sparse adaptive graph convolutional unit with a prior knowledge constraint was proposed to perform adaptive spatial modeling of physical and logical dependency for skeleton sequences, thus achieving the sparse modeling of the discriminative spatial relationships. Subsequently, a temporal context module was introduced to construct the remote context dependency in the temporal dimension, hence determining the global changes of the task. A multi-domain attention learning module was also developed to integrate the static spatial features and dynamic temporal features, and then to emphasize the salient feature selection in the channel domain, thereby capturing the multi-domain fine-grained information. Finally, the evaluation results using a dataset with 148 patients and 870 samples confirmed the effectiveness and reliability of our scheme, and the method outperformed other related state-of-the-art methods. Our contactless method provides a new potential tool for automated PD assessment and telemedicine.

Alzheimer's Disease Classification With a Cascade Neural Network

Classification of Alzheimer's Disease (AD) has been becoming a hot issue along with the rapidly increasing number of patients. This task remains tremendously challenging due to the limited data and the difficulties in detecting mild cognitive impairment (MCI). Existing methods use gait [or EEG (electroencephalogram)] data only to tackle this task. Although the gait data acquisition procedure is cheap and simple, the methods relying on gait data often fail to detect the slight difference between MCI and AD. The methods that use EEG data can detect the difference more precisely, but collecting EEG data from both HC (health controls) and patients is very time-consuming. More critically, these methods often convert EEG records into the frequency domain and thus inevitably lose the spatial and temporal information, which is essential to capture the connectivity and synchronization among different brain regions. This paper proposes a cascade neural network with two steps to achieve a faster and more accurate AD classification by exploiting gait and EEG data simultaneously. In the first step, we propose attention-based spatial temporal graph convolutional networks to extract the features from the skeleton sequences (i.e., gait) captured by Kinect (a commonly used sensor) to distinguish between HC and patients. In the second step, we propose spatial temporal convolutional networks to fully exploit the spatial and temporal information of EEG data and classify the patients into MCI or AD eventually. We collect gait and EEG data from 35 cognitively health controls, 35 MCI, and 17 AD patients to evaluate our proposed method. Experimental results show that our method significantly outperforms other AD diagnosis methods (91.07 vs. 68.18%) in the three-way AD classification task (HC, MCI, and AD). Moreover, we empirically found that the lower body and right upper limb are more important for the early diagnosis of AD than other body parts. We believe this interesting finding can be helpful for clinical researches.

Skeleton edge motion networks for human action recognition

Human skeleton is receiving increasing attention from the community of human action recognition due to its robustness to complex image backgrounds. Previous methods usually utilize body joint-based representation, i.e., joint locations, while leaving edge-based movement poorly investigated. In this paper, we propose a new human action recognition method, skeleton edge motion networks (SEMN), to further explore the motion information of human body parts. Specifically, we address the movement of skeleton edge by using the angle changes of skeleton edge and the movement of the corresponding body joints. We then devise the proposed skeleton edge motion networks by stacking multiple spatial-temporal blocks to learn a robust deep representation from skeleton sequences. Furthermore, we propose a new progressive ranking loss to help the proposed skeleton edge motion networks maintain temporal order information in a self-supervised manner. Experimental results on five popular human action recognition datasets, PennAction, UTD-MHAD, NTU RGB+D, NTU RGB+D 120, and CSL, demonstrate the effectiveness of the proposed method.

Learning Spatial-Temporal Representations Over Walking Tracklet for Long-Term Person Re-Identification in the Wild

Long-term person re-identification (re-ID) aims to build identity correspondence of the Target Subject of Interest (TSI) exposed under surveillance cameras over a long time interval. Compared to the conventional short-term re-ID studied by most existing works, it suffers an additional problem: significant dressing change observed with time lapsing. Unfortunately, this variation in long-term person re-ID case contradicts the assumption of prior short-term re-ID approaches, and thus causes significant difficulties if conventional short-term re-ID methods are applied. To address the problem, this paper proposes to learn hybrid feature representation via a two-stream network named SpTSkM, including a spatial-temporal stream and a skeleton motion stream. The former performs directly on image sequences, which tends to learn identity-related spatial-temporal patterns such as body geometric structure and body movement. The latter operates on normalized 3D skeletons by adapting graph convolutional network, which tends to learn pure motion patterns from skeleton sequences. Both streams extract fine-grained level time-gap stable information that is robust to appearance changes in long-term re-ID and meanwhile maintains sufficient discriminability to differentiate different people. The final matching metric is obtained by mixing information of the two streams in a score-level fusion strategy. In addition, we collect a Cloth-Varying vIDeo re-ID (CVID-reID) dataset particularly for long-term re-ID. It contains video tracklets of celebrities posted on the Internet. These videos are snapshots under extremely different scenarios that include highly dynamic background, diverse camera views and abundant cloth variations on each TSI. These factors cause CVID-reID more complicated and closer to practice. Our experiments demonstrate the difficulty of long-term person re-ID and also validate the effectiveness of the proposed SpTSkM, showing the best performance.

A New Method for Mapping Active Joint Locations of Skeletons to Pre-Shape Space for Action Recognition

Being a class of effective feature descriptors for action recognition, action representations based on skeleton sequences have yielded excellent recognition results. Most methods used to construct these action representations are based on the information from all the joint positions in actions. Unfortunately, some joints in the actions do not improve the accuracy of action recognition, and may even cause unnecessary inter-class errors. In this study, the authors propose a new method for action recognition by selecting active joints which are closely related to the movement of the body as the first step. Further, a skeleton is characterized as a set of its active-joint positions, and the set can be mapped to a point on pre-shape space to filter out the scale and translation variability. Then, a skeleton sequence (an action) can be regarded as points on the space. Because the timing-sequence relationship between skeletons is very valuable for action recognition, a tensor-based linear dynamical system (tLDS) is employed to model the temporal information of the action. To avoid using a finite-order sequence to estimate the infinite-order feature descriptor of a tLDS, the descriptor is mapped to a point on an infinite Grassmannian composed of the extended observability subspaces. The action is classified using sparse coding and dictionary learning (SCDL) on the infinite Grassmannian. Experimental results demonstrate that the recognition accuracies of the proposed method outperform state-of-the-art ones on four different action datasets.

View transform graph attention recurrent networks for skeleton-based action recognition

Human action recognition based on skeleton recently has attracted attention of researchers due to the accessibility and popularity of the 3D skeleton data. However, it is complicated to effectively represent spatial–temporal skeleton sequences given the large variations of action representations when they are captured from different viewpoints. In order to get a better representation of the spatial–temporal skeletal features, this paper introduces a view transform graph attention recurrent networks (VT+GARN) method for view-invariant human action recognition. We design a view-invariant transform strategy based on the sequence to reduce the influence of different views on the spatial–temporal position of skeleton joint. Then, the graph attention recurrent network automatically calculates the coefficient of attention and learns the representation of spatiotemporal skeletal features after the transformation and outputs the classification result. Ablation studies and extensive experiments on three challenging datasets, Northwestern-UCLA, NTU RGB+D and UWA3DII, demonstrate the effectiveness and superiority of our method

Spatial–temporal graph attention networks for skeleton-based action recognition

Human action recognition based on skeleton currently has attracted a wide range of attention. The structure of skeleton data exists in the form of graph, thus most researchers use graph convolutional networks (GCN) to model skeleton sequences. However, the graph convolution network shares the same weight for all neighbor nodes and relies on the connection of graph edges. We introduce a method, a spatial–temporal graph attention networks (ST-GAT), to overcome the disadvantages of GCN. First, the ST-GAT defines the spatial–temporal neighbor nodes of the root node and the aggregation function through the attention mechanism. The adjacency matrix is only used in GAT to define related nodes, and the calculation of association weight is dependent on the feature expression of nodes. Then ST-GAT network attaches the obtained attention coefficient to each neighbor node to automatically learn the representation of spatiotemporal skeletal features and output the classification results. Extensive experiments on two challenging datasets consistently demonstrate the superiority of our method.

Fuzzy Integral-Based CNN Classifier Fusion for 3D Skeleton Action Recognition

Action recognition based on skeleton key joints has gained popularity due to its cost effectiveness and low complexity. Existing Convolutional Neural Network (CNN) based models mostly fail to capture various aspects of the skeleton sequence. To this end, four feature representations, which capture complementary characteristics of the sequence of key joints, are extracted with novel contribution of features estimated from angular information, and kinematics of the human actions. Single channel grayscale images are used to encode these features for classification using four CNNs, with the complementary nature verified through Kullback-Leibler (KL) and Jensen-Shannon (JS) divergences. As opposed to straightforward classifier combination generally used in existing literature, fuzzy fusion through the Choquet integral leverages the degree of uncertainty of decision scores obtained from four CNNs. Experimental results support the efficacy of fuzzy combination of CNNs to adaptively generate final decision score based upon confidence of each information source. Impressive results on the challenging UTD-MHAD, HDM05, G3D, and NTU RGB+D 60 and 120 datasets demonstrate the effectiveness of the proposed method. The source code for our method is available at https://github.com/theavicaster/fuzzy-integral-cnn-fusion-3d-har

Research on Discriminative Skeleton-Based Action Recognition in Spatiotemporal Fusion and Human-Robot Interaction

A novel posture motion-based spatiotemporal fused graph convolutional network (PM-STGCN) is presented for skeleton-based action recognition. Existing methods on skeleton-based action recognition focus on independently calculating the joint information in single frame and motion information of joints between adjacent frames from the human body skeleton structure and then combine the classification results. However, that does not take into consideration of the complicated temporal and spatial relationship of the human body action sequence, so they are not very efficient in distinguishing similar actions. In this work, we enhance the ability of distinguishing similar actions by focusing on spatiotemporal fusion and adaptive feature extraction for high discrimination information. Firstly, the local posture motion-based attention (LPM-TAM) module is proposed for the purpose of suppressing the skeleton sequence data with a low amount of motion in the temporal domain, and the representation of motion posture features is concentrated. Besides, the local posture motion-based channel attention module (LPM-CAM) is introduced to make use of the strongly discriminative representation between different action classes of similarity. Finally, the posture motion-based spatiotemporal fusion (PM-STF) module is constructed which fuses the spatiotemporal skeleton data by filtering out the low-information sequence and enhances the posture motion features adaptively with high discrimination. Extensive experiments have been conducted, and the results demonstrate that the proposed model is superior to the commonly used action recognition methods. The designed human-robot interaction system based on action recognition has competitive performance compared with the speech interaction system.

Deep learning-based multi-modal approach using RGB and skeleton sequences for human activity recognition

The deep learning techniques have achieved great success in the application of human activity recognition (HAR). In this paper, we propose a technique for HAR that utilizes the RGB and skeleton information with the help of a convolutional neural network (Convnet) and long short-term memory (LSTM) as a recurrent neural network (RNN). The proposed method has two parts: first, motion representation images like motion history image (MHI) and motion energy image (MEI) have been created from the RGB videos. The convnet has been trained, using these images with feature-level fusion. Second, the skeleton data have been utilized with a proposed algorithm that develops skeleton intensity images, for three views (top, front and side). Each view is first analyzed by a convnet, that generates the set of feature maps, which are fused for further analysis. On top of convnet sub-networks, LSTM has been used to exploit the temporal dependency. The softmax scores from these two independent parts are later combined at the decision level. Apart from the given approach for HAR, this paper also presents a strategy that utilizes the concept of cyclic learning rate to develop a multi-modal neural network by training the model only once to make the system more efficient. The suggested approach privileges for the perfect utilization of RGB and skeleton data available from an RGB-D sensor. The proposed approach has been tested on three famous and challenging multimodal datasets which are UTD-MHAD, CAD-60 and NTU-RGB + D120. Results have shown that the stated method gives a satisfactory result as compared to the other state-of-the-art systems.

Human Skeleton Data Augmentation for Person Identification over Deep Neural Network

With the advancement in pose estimation techniques, skeleton-based person identification has recently received considerable attention in many applications. In this study, a skeleton-based person identification method using a deep neural network (DNN) is investigated. In this method, anthropometric features extracted from the human skeleton sequence are used as the input to the DNN. However, training the DNN with insufficient training datasets makes the network unstable and may lead to overfitting during the training phase, causing significant performance degradation in the testing phase. To cope with a shortage in the dataset, we investigate novel data augmentation for skeleton-based person identification by utilizing the bilateral symmetry of the human body. To achieve this, augmented vectors are generated by sharing the anthropometric features extracted from one side of the human body with the other and vice versa. Thereby, the total number of anthropometric feature vectors is increased by 256 times, which enables the DNN to be trained while avoiding overfitting. The simulation results demonstrate that the average accuracy of person identification is remarkably improved up to 100% based on the augmentation on public datasets.

Skeleton-based action recognition with hierarchical spatial reasoning and temporal stack learning network

Skeleton-based action recognition aims to recognize human actions by exploring the inherent characteristics from the given skeleton sequences and has attracted far more attention due to its great important potentials in practical applications. Previous methods have illustrated that learning discriminative spatial and temporal features from the skeleton sequences is a crucial factor to recognize human actions. Nevertheless, how to model spatio-temporal evolutions is still a challenging problem. In this work, we propose a novel model with hierarchical spatial reasoning and temporal stack learning network (HSR-TSL) to explore the discriminative spatial and temporal features for human action recognition, which consists of a hierarchical spatial reasoning network (HSRN) and a temporal stack learning network (TSLN). Specifically, the HSRN employs a hierarchical residual graph neural network to capture two-level spatial features: intra spatial information of each part and body-level structural information between each part. The TSLN models the detailed temporal dynamics of skeleton sequences by a composition of multiple skip-clip LSTMs. During training, we develop a clip-based incremental loss to effectively optimize the model. We perform extensive experiments on five challenging benchmarks to verify the effectiveness of each component of our model. The comparison results illustrate that our approach significantly boosts the performances for skeleton-based action recognition.

S3D-CNN: skeleton-based 3D consecutive-low-pooling neural network for fall detection

Most existing deep-learning-based fall detection methods use either 2D neural network without considering movement representation sequences, or whole sequences instead of only those in the fall period. These characteristics result in inaccurate extraction of human action features and failure to detect falls due to background interferences or activity representation beyond the fall period. To alleviate these problems, a skeleton-based 3D consecutive-low-pooling neural network (S3D-CNN) for fall detection is proposed in this paper. In the S3D-CNN, an activity feature clustering selector is designed to extract the skeleton representation in depth videos using pose estimation algorithm and form optimized skeleton sequence of fall period. A 3D consecutive-low-pooling (3D-CLP) neural network is proposed to process these representation sequences by improving network in terms of layer number, pooling kernel size, and single input frame number. The proposed method is evaluated on public and self-collected datasets respectively, outperforming the existing methods.

Spatio-Temporal Reconstruction for 3D Motion Recovery

This paper addresses the challenge of 3D motion recovery by exploiting the spatio-temporal correlations of corrupted 3D skeleton sequences. We propose a new 3D motion recovery method using spatio-temporal reconstruction, which uses joint low-rank and sparse priors to exploit temporal correlation and an isometric constraint for spatial correlation. The proposed model is formulated as a constrained optimization problem, which is efficiently solved by the augmented Lagrangian method with a Gauss–Newton solver for the subproblem of isometric optimization. The experimental results on the CMU motion capture dataset, Edinburgh dataset, and two Kinect datasets demonstrate that the proposed approach achieves better motion recovery than the state-of-the-art methods. The proposed method is applicable to Kinect-like skeleton tracking devices and pose estimation methods that cannot provide accurate estimation of complex motions, especially in the presence of occlusion.